Anda di halaman 1dari 24

Fortnightly Thoughts

February 18, 2015 Issue 85

The real consequences of artificial intelligence

From the editor: In this edition we cogitate on the increasing transition of artificial intelligence
into the mainstream, commercial world and its potentially disruptive implications and uses. We
interview three experts on this subject and include our global analysts views.
The substantial progress made over the tentacles reaching healthcare and AI can also instigate disruption in industries
last decade in the capabilities and cost of manufacturing as well as logistics and much like the advent of the internet did.
parallel computing, algorithms, big data and energy consumption. The ability to think While we do not foresee the indiscriminate
the move to the cloud is set to bring and learn enables AI-aided technologies to demise of industry incumbents, we expect
artificial intelligence out of labs and into the constantly improve and refine and when AI to act as a necessary stay-in-the-game
real, mainstream world. This has applied to decision making in businesses cost to incumbents and also open the door
implications for information-intensive we believe this can lead to better cost and to capital and labor-light new entrants.
sectors as well as businesses that rely on capital allocation, lower error rates and However AI can reinforce dominance in
highly skilled personnel. AI is already accelerated innovation. AI is aimed at industry leaders that have a pre-existing
driving changes in advertising augmenting human decision making, but in edge in terms of access to proprietary data.
(programmatic ad buying), parts of retail many areas it could replace humans or Apart from them, we see opportunity in AI
(customised recommendations) and resolve the shortage of human skills (think enablers such as NEC, Nidec, Verint and
investing. With other sectors becoming collaborative robots in warehouses and Criteo, as well as users of collaborative
increasingly data-intensive we see its hospitals). robots and AI like Amazon and Baidu.

The age of innovation Whats inside

Total patent grants by the USPTO by field of technology
The real consequences: Our lead article on AI 2
Computer Graphics Processing
Interview withRaj Rajkumar: Prof., Carnegie 7
Image Analysis Mellon University
4000 Database Management* The road to automation: Takafumi Hara on the 9
Financial, Business Practice*
future of connected vehicles

Vehicles, Navigation* Interview withRodney Brooks: Founder, 11

2000 chairman and CTO of Rethink Robotics
Speech Processing,
1000 The start of the collaboration era: Yuichiro 13
Artificial Intelligence *
Isayama on collaborative robots
0 Virtual Machine Task
Interview withManoj Saxena: Founding General 15














Partner, The Entrepreneurs Fund

Note:all *marked categories relate to data processing; Financial, Business
Practice includes things like cost/price determination
Source: USPTO.
AI is this time different?: Greg Dunham on the 17
use of big data in AI applications
Hugo Scott-Gall Sumana Manohar, CFA The age of robo advisors: Alexander Blostein 18
+1 (212) 902 0159 +44 (20) 7051 9677 on automated investing
Goldman, Sachs & Co. Goldman Sachs International

Goldman Sachs does and seeks to do business with companies covered in its research reports. As a result, investors should be
aware that the firm may have a conflict of interest that could affect the objectivity of this report. Investors should consider this
report as only a single factor in making their investment decision. For Reg AC certification and other important disclosures, see
the Disclosure Appendix, or go to Analysts employed by non-US affiliates are not
registered/qualified as research analysts with FINRA in the U.S.

The Goldman Sachs Group, Inc. Goldman Sachs Global Investment Research
Fortnightly Thoughts Issue 85

The real consequences of AI

A broad way to define artificial intelligence or AI is that it is any greater usage and so on. Taken together, the improved capability
intelligence exhibited by machines or software. By this we mean of machines to collate and comprehend more information and
machines which can think, which are able to not just process data better optimise results is driving the broader and more significant
into information, but also derive knowledge from that information to potential for AI.
augment human decision making or to act independently. It is
difficult to draw a line between AI and a smart piece of code or Stores of value
smart connected devices, but the key difference here is that AI Stored data in the US by industry, petabytes
refers to technology that can be taught or is capable of self-learning Discrete manufacturing
and so can continually improve itself; capabilities once thought to Govt.
be the forte of human beings. Take the case of Google search, Comm & media

which Nick Bostrom, Professor at Oxford University and a leading Process manufacturing
thinker in this area, highlights as the best example of AI to date in
Healthcare providers
his book Superintelligence. What makes the original Google Securities/investment svcs.
algorithm truly powerful is that it is capable of learning from millions Professional svcs.
of users searching, ignoring and clicking through results each day, Retail

teaching it to yield better results every time. AI can similarly be Education

applied to continually refine decision making in any information-
driven business model to streamline costs, enable more efficient Wholesale
allocation of resources, improve product quality and accelerate Utilities
innovation. This means that AI can be deployed in sectors that rely Resources

on skilled employees (manufacturing, software, engineering), Consumer/ recreation

skilled intermediaries (doctors for pharma, mechanics for cars) or
0 100 200 300 400 500 600 700 800 900 1000
skilled users (autos, agriculture), and even in operations and
Source: IDC; BLS; McKinsey Global Institute analysis
processes that are simple and mundane. We see AI as the next leg
of the technological revolution that digitises decision making and as
a result threatens swathes of human expertise and pattern Where is AI now and where can it go?
recognition skills, lowers entry barriers for new competitors and AI going mainstream has broad implications, but it is currently most
adds to the disinflationary forces in the world. evident in data-heavy sectors. Algorithms behind Spotify, Netflix
and Amazon are all aimed at driving greater customisation and user
The stars are aligned engagement. Retailers, both online and offline, are increasingly
Artificial intelligence is by no means a new concept; however, even deploying AI to leverage their customer interaction data to boost
almost 60 years after the term was first coined it continues to revenue at lower selling costs. i.e. customised loyalty card-based
conjure a perception that it is something that will be realised in the promotions could become even more sophisticated. Smarter
far future. This is the paradox of the AI effect as soon as artificial algorithms are also automating the process of buying and bidding
intelligence achieves a commercial application, it ceases to be for online ad space; programmatic ad buying, enabled by
viewed as AI and is instead considered to be a really clever piece companies like Rocket Fuel, The Rubicon Project and Criteo, is
of code or a smart device. But in reality, AI is alive and kicking, and reshaping digital advertising, driving cost efficiency for advertisers
its influence is already evident in many industries, which we delve and publishers. And in the world of investing, algorithmic trading
into in the next few paragraphs. However, we are writing about AI has been a force for the past few years, thanks mostly to its speed
now because we think that it is on the cusp of a period of more advantage. But more recent improvements in parallel computing
rapid growth in its use and applications. The reasons are multi- technology have resulted in more powerful tools that are helping
pronged, but the improvements in the capability and economics of analysts digest reams of unstructured data, helping them replicate
hardware is a good place to start. On one hand, many different years of experience and acquired pattern recognition skills (see
device components such as sensors, cameras, radars, lasers etc. page 18 for Alex Blostein on robo-advisors). Private firm Kensho is
have become much cheaper in recent years thanks mostly to the an example here and so is BlackRocks Aladdin (on understanding
smartphone revolution, which has driven innovation and economies investment risks).
of scale at an unprecedented pace. This has resulted in more
connected devices and in turn, a lot more data being available for Apples Siri, Amazons Alexa (inside Echo), Microsofts Cortana and
analysis. On the processing side, significant progress made in the Google Now are also examples of AI-aided personal assistants on
field of neural networks and parallel computing in the last decade our devices. And this is perhaps one use case of AI which has
has meant that computers are now better able to understand even grabbed the most mindshare of everyday consumers, by making
unstructured data, like oral conversations and pictures. that inconspicuous, but huge shift between text and voice
commands and between reactive and pre-emptive search results
Deep mines and recommendations. This is the difference between searching for
At the same time, the advent of deep learning, only about a decade weather forecasts on smartphones versus them pre-emptively
old in its current form, has enabled faster and more accurate advising users to take an umbrella to their appointment. As they
reasoning and processing algorithms in most of todays AI tools. learn more and improve further, it is easy to imagine devices
Plus, as Manoj Saxena argues on page 15, the move to the cloud answering queries before they are asked. It is interesting to note
has meant that massive computing capacity and near infinite here that last year, Amazon was granted a patent for anticipatory
processing power is now available at very low prices and requires shipping, which aims to prepare a package for delivery before the
no upfront installation or capital spend, making AI more accessible customer actually makes the purchase, based on his or her
to businesses than it has ever been. This should contribute to the browsing and purchasing behaviour.
network effect that accelerates the capability and adoption of AI
the more it is used, the better it becomes which should result in

Goldman Sachs Global Investment Research 2

Fortnightly Thoughts Issue 85

Patently important employee) include media, financial services and technology. All
Patent grants by the USPTO office by technology class and owner said, artificial intelligence should augment human intelligence and
Artificial Intelligence * Computer Graphics Processing lead to better-informed decisions, and depending on the nature of
Image Analysis Vehicles, Navigation* the industry it is deployed in, it can be disruptive to incumbents or
6,000 Design & Simulation* Speech Processing, Translation*
Financial, Business Practice* Database Management* at the very least it can re-allocate industry profits among them.
Virtual Machine Task
It is useful here to draw parallels between AI and other
4,000 technological waves the use of plastic money, automation and
even the advent of e-commerce. In the latter case, the investment
3,000 opportunities were quite varied in the early years of online adoption.
Whether it was in media, apparel or grocers, new entrants were
disruptive to existing market shares, while amongst the incumbents,
there was a marked difference in the performance of those who
were willing and able to adapt early and those that were less
0 capable of making the shift (especially those tied down by legacy
fixed assets). Today, most large retailers have established an










online presence and competition in some of these sectors is

reverting back to the product quality, price or range (old school
Note:all *marked categories relate to data processing; Financial, Business
Practice includes things like Cost/Price Determination competition), rather than the convenience of the channel, but that
Source: USPTO. has followed a costly and painful transition period. The longer the
transition period, the more persistent the shifts in market share are
Which industries are becoming more information-intensive? (think online impact on food retail versus classifieds), even though
Healthcare, and particularly pharmaceuticals, stands out here. One the technology ultimately becomes part and parcel of the business.
of the most well-known examples of cognitive computing of course AI we think is headed in the same direction. In other words, AI can
is IBMs Watson, which won Jeopardy! in 2011. Its first commercial introduce new pure-play entrants and re-shape the revenue pool,
foray was in healthcare in 2013, helping doctors lower the error rate until it is eventually adopted more broadly in the industry (much like
in cancer diagnoses. As a result of the significant progress made in Netflix vs. other cable companies adopting OTT, Tesla vs. other
gene sequencing over the past decade, healthcare providers now OEMs building EVs). It will become critical for competition but
have access to a substantial and rapidly growing amount of genetic irrelevant for strategy, which means that the advantages of
data; and given that the true value in genomics lies in the ability to adopting AI are easily outweighed by the risks of not using it. And
digest and understand this information to translate it into more so, investment opportunities are more likely to be found by
personalised and accurate treatment, this is another area where AI identifying AI tool providers, new entrants and incumbents who are
can play a huge role. Education, energy-consumption, logistics and best suited to adapt or respond.
manufacturing are all becoming more information-driven as well
and hence present opportunities for AI deployment. Who provides the tools? Within industries which have been early
adopters of AI, there are few listed companies that provide specific
tools like advertising and programmatic ad tech vendors or financial
Automatic genomics
Data output measured by number of terabases analytics. Japans NEC is dominant in face recognition and text
analysis, which has broad applications in areas such as security
and marketing. Tech firms that provide advanced data analytical
and visualisation tools, like Verint Systems and Marketo are
2,000 similarly exposed to this space (see longer list on page 6). In terms
of general artificial intelligence, IBMs Watson remains one of the
most prevalent providers of AI solutions. It now offers 13 different
services including speech-to-text translation and trade-off analytics
on Bluemix, its cloud platform for developers. IBMs CEO has
previously mentioned that she hopes that the company will
generate $10 bn in revenues from Watson in the next decade
(versus $100 mn as of October 2014). Having said that, our
500 analysts dont foresee a significant revenue contribution from
Watson in the near-term. But apart from these, most solution
providers remain private and relatively small, reflecting the nascent
2009 2010 2011 2012 2013 2014e
nature of this space. Greg Dunham has more on this on page 17.
Source: The Broad Institute
Having said that, the list of AI start-ups that have been acquired by
established companies in tech and other sectors has been growing
AI can also lower the reliance on high-skilled personnel, especially rapidly in recent years. Google has been particularly acquisitive,
in sectors that face a shortage of talent, thus helping companies buying deep learning, image recognition and neural network
counter wage inflation and make their product or service available technologies. Yahoo, Microsoft, Intel, LinkedIn, Walmart, and
to a wider audience. Again healthcare is an area where this is Infosys have similarly invested in this field. This is of course one
relevant; AI could go some way in resolving the shortage of care
workers, nurses and physicians. Other sectors that are both labour-
intensive (wages as a % of costs) and talent-intensive (wages per

Goldman Sachs Global Investment Research 3

Fortnightly Thoughts Issue 85

M&A activities related to Artificial Intelligence

Acquirer Target Businessdescription Acquirer Target Businessdescription
Technology Sociamedia
Google DeepMind($400m)2014 Deeplearningtounderstandimages,textandvideos Facebook Wit.ai2015 SirilikeAPIvoiceinterface
DarkBlue2014 Deeplearningtounderstandnaturallanguage Twitter Madbits2014 Imageunderstandingusingdeeplearning
VisionFactory2014 Visualrecognitionusingdeeplearning Pinterest VisualGraph2014 Machinevision,imagerecognition,visualsearch
Jetpac2014 Imagerecognitionandneuralnetworktechnology LinkedIn Bright($120m)2014 AIandbigdataalgorithimstoconnectusers
DNNresearch2013 Deepneuralnetworks,languageprocessing Ecommerce
Yahoo IQEngines2013 Imagerecognitiontotagandorganisephotos Ebay Apptek2014 Hybridmachinetranslationusingmachinelearning
LookFlow2013 Enhancedimagerecognition Amazon EviTechnology2013 Internetsearchandvoicerecognition
SkyPhrase2013 Naturallanguageprocessingtechnology Kiva($775m)2012 Manufacturingmobileroboticfulfillmentsystem
IBM Kenexa($1.3bn)2012 Dataanalysistohelprecruitandretainworkers DeutscheTel Magisto2014 AIvideostorytellingplatform(partnership)
Microsoft Equivio2015 Machinelearningtechnologyforinfogovernance Retail
RevolutionAnalytics2015 Statisticalcomputing,predictiveanalytics Staples Runa2013 Specialistinecommercepersonalizationtech
Infosys Panaya($200m)2015 Automatedcloudbasedqualitymanagementservices Walmart Luvocracy2014 Discoveringrecommendationsbyfriends
Dropbox AnchoviLabs2012 ImageclassificationusingAI InkiruInc.2013 Predictiveanalyticsapplications
Intel Indisys2013 Naturallanguagerecognition HomeDepot BlackLocus2012 DataanalyticsinnovationlabbasedinAustinTX
Omek($40m)2013 Makerofgesturebasedinterfaces Capgoods
Stryker MAKOSurgical($1.65b)2013 Advanceroboticassistedsurgeryinorthopedics GE Pivotal($100m+)2013 Developsdataanalyticsofferings
Monsanto ClimateCorp.($930m)2013 Underwritingweatherinsuranceinrealtime Schneider InStepSoftware2014 Realtimeperformancemgt,predictiveanalytics
Source: Goldman Sachs Global Investment Research

route that incumbents can take to adopt AI, the other being that waste and applying glue). On page 11, we interview Rodney
they develop AI capabilities in-house (like social media companies Brooks of Rethink Robotics, which builds the Baxter robot and he
for instance). But more broadly, incumbents that are best suited to notes that collaborative robots, which are smaller, mobile and often
respond to any AI-led disruption are likely to be those that have interactive, are better suited for small manufacturers with limited
access to proprietary data. With AI, data is only set to become a budgets, staff and space, rather than being disruptive to industrial
stronger entry barrier. A large and growing population dataset robots that perform heavy and complex tasks. Another application
allows the machine or the software to learn faster and deeper. In for collaborative robotics is in logistics and warehouses; Amazon
other words, more data can make a clever algorithm cleverer and acquired Kiva Systems in 2013 which provides automation
so an early mover with a unique or large dataset can build a huge solutions in its fulfilment centres, using mobile robots and
advantage. AI can thus reinforce dominance in industries where sophisticated control technology to identify and retrieve packages
some existing players have a pre-existing data advantage, while for that have been ordered, allowing for faster cycle times, longer
the others, it can prove to be quite disruptive. operational hours and reduced labour requirements. Again, these
are simple, but cumbersome and tedious jobs.
Lets start something special
Artificial Intelligence start-up funding At your service
ArtificialIntelligence Totalraised Sale of service robots by type
Startup ($m) StartupDescription 2013 Average annual sales, 2014-17E

Senti entTechnol ogi e 144 Da ta a na l ys i s throughma s s i vel ys ca l edAI Defence applications

Rethi nkRoboti cs 127 Robots forproducti ona ndres ea rch Milking Robots
Vi ca ri ous 72 Adva ncedi ma gerecogni ti on Logistic systems
Moderni zi ngMedi ci n 55 Cl oudba s edEl ectroni cMedi ca l As s i s ta nt Automated Guided Vehicles
ContextRel eva nt 44 Us i ngmachi nel ea rni ngtoa na l ys ebi gda ta
Medical robots
Na rrati veSci ence 32 Generatewri ttenna rati ves fromda ta
Mobile platforms
Scal edInference 27 Pl atformforgenera l purpos eAI
Elderly and handicap assistance
Kens ho 26 Automa ti onoffi nanci al res ea rch
Construction, demolition
Bui l di ngIQ 21 Energyma nagements oftwa repl a tform
Professional cleaning
ExpectLa bs 15 Voi cea cti va teddi s covery
Inspection, maintenance
Bl ueRi verTechnol og 13 Lettucethi nni ngus i ngma chi nel ea rni ng
Underwater systems
Na ra Logi cs 13 Bi gdata a nal ys i s
Rescue and security
TempoAI 13 Mobi l eproducti vi tyapporga ni s es us er's da y
PR (guides in supermarkets, museums etc.)
Sentri a n 12 Remotebi os ens ors topreventhos pita l i s ation
0 2000 4000 6000 8000 10000 12000 14000 16000
Preci s i onHa wk 11 Aeri al da ta a na l ys i s ,AIs oftwareforUAVs
Note: 2013 data not available for PR robots, only handicap assistance sales
Meta Mi nd 8 Na tura l l a ngua ge&i ma gerecogni ti on
considered for 2013 in the elderly and handicap assistance category
Source: Company data and News Sources Source: IFR World Robotics

Baxter, is that you? Helping the age old problem

So far we havent spent a lot of time on the hardware aspect of Outside of factories and warehouses, the potential for collaborative
machine intelligence output because it is not paramount for AI. AI is and service robots is perhaps highest in the field of healthcare
the brain inside a robot, a robot is the container for AI and each can ageing demographics and a shortage of care workers is an
exist without the other. In previous issues of Fortnightly Thoughts, emerging issue in many developed markets and particularly in
we have written at length about smart, flexible robots becoming Japan. This is one of the reasons that in 2013, Prime Minister
more adept and expanding their scope beyond traditionally Shinzo Abe's government announced subsidies to encourage
automated end-markets (like machine tool, auto manufacturing), companies to develop robotic care for the elderly (the government
into previously labour-intensive sectors like consumer electronics forecasts the nursing care robotic market to grow from $140 mn in
and food production. Robots equipped with AI can further negate 2015 to $3.4 bn by 2035). These are robots that can understand
the need for human instruction and intervention and, counter- human conversations, avoid collisions, lift heavy objects (including
intuitively, this is likely to have the greatest impact in tasks that are people) and in general, make life incrementally easier for users.
mundane, simple and time-consuming (like cleaning tools, moving

Goldman Sachs Global Investment Research 4

Fortnightly Thoughts Issue 85

Toyota and Panasonic are among the companies focused on Ifs and bots
building collaborative robots for hospitals and elderly care at home. Share of article edits and creations by bots on Wikipedia
Share of edits by bots Share of creations by bots

And smart machines do not necessarily have to be robots. Hearing 80%

aids, made by William Demant, Sonova and GN Store Nord, have 70%
increasingly become more intelligent, with the devices now able to
recognize what type of environment the user is in, i.e. a loud
restaurant, an office, or in the car and listening to music. Upon 50%

distinguishing the environment, hearing aids automatically change 40%

the hearing program, and utilize processing algorithms which are
best suited to that specific setting. For instance, the hearing aids
may lower the volume of the background noise in the restaurant, 20%

and focus the microphones on the person sitting across the table, 10%
so that the user can hear the conversation clearly, but hears a lot
less of the clinking plates and other conversations in the restaurant. 0%








All languages


Look no hands
Another sector where machine intelligence has been attracting a lot Note: Only registered bots have been included. Only article namespace(s) have
of attention in recent times is automobiles. There is no denying that been counted.
cars are becoming more tech-intensive, with connected cars Source: Wikipedia
gradually replacing the existing stock of vehicles (see Takafumi
Hara on page 9 for more on this). But what about fully autonomous So, what is everybody going to do?
cars? We interview Prof. Raj Rajkumar from Carnegie Mellon on We have asked this question many, many times before. And we
page 7, who expects the technology to become feasible by 2020, keep coming back to this because each year the capability and
thanks to the falling costs of hardware components actuators, reach of technology seems to increase substantially, strengthening
sensors, cameras, scanners, etc., increasing connectivity and the advantage of digital labour versus human labour, both in terms
improving capacity of the software within these vehicles. of cost and capability. Machines already provide the advantage of
Autonomous cars can make roads safer, reduce energy speed, precision and endurance (they dont need to go to sleep,
consumption, make the product more accessible (to older they dont get stressed or tired). With AI, if we can combine these
consumers, those who cannot drive), reduce the costs of ownership advantages with breadth and expertise, including pattern
and eventually lower the aggregate demand, by optimising car recognition and a better contextual understanding for decision
usage. A car is a unique example of machine intelligence because making (crucial in the arsenal of skilled professionals), then
its one of the very few consumer goods that requires a license to machines can compete against human labour in a wider range of
operate, given that driving is one of the most dangerous and skill- jobs. Most AI applications being developed today are aimed at
intensive operations that most of us perform on a regular basis. augmenting, rather than replacing, human skill. But that still means
that AI is increasing the supply of expertise, thus reducing its value
This brings us back to the question of skill and the relative in people.
economics of digital versus human labour. We have previously
Of course there are counter arguments to this; many experts argue
written about how the internet is disrupting businesses that rely on
that artificial intelligence is still no match for natural stupidity. While
information asymmetry like travel agencies and real estate rentals.
machines have been able to replicate human tasks that require
AI, we think, can take this further and bring about a step change in
thinking, many have failed to perform simpler jobs that dont require
the economics of business models and professions that rely on
much thought. And AI could also be limited by regulation that seeks
knowledge asymmetry. What we are referring to here are industries
to save jobs or protect privacy; if companies are restricted from
where just the availability of information isnt sufficient, but
collecting or using some of their data, AI-based business models
expertise and knowledge are requisite (Google can provide a
are harder to imagine. It is also not difficult to imagine consumer
diagnosis based on our symptoms, but we would still prefer to see
reluctance to completely trust machines to drive their cars or
a doctor or we can refer to a recipe, but a chef can still prepare a
prescribe medical treatments. So it is unlikely that machines will
dish better than an average cook). It is in these areas where AI
replace human expertise and labour drastically in the near future,
could eventually make a big difference. IBMs Watson for instance
but we do expect this to continue to weigh on wages.
has been deployed by lawyers to scrutinise thousands of legal
statutes and documents to answer legal questions and soon it All said, we find AI both gloomy and exciting in equal measure. It is
could do the same for consumers. That is a cheaper and faster the next phase of the data and tech revolution, as the world
alternative for paralegals or lawyers, with years of experience and becomes more tech-intensive and machines become even more
know-how. Article writing is similarly getting automated for data- capable. More specifically, AI is the digitisation of decision making
heavy news reporting. The company Automated Insights for and that can lead to varied consequences in different sectors.
example churned out more than 4000 corporate earnings stories Below, we have a list of companies from our global coverage
per quarter for the Associated Press last year, up from 300 that AP exposed to the broad field of AI and machine intelligence, and there
could previously produce manually. AI can similarly be deployed in is one final observation wed like to make before we leave you. The
augmenting human talent in a broad range of industries; like assist table includes AI enablers, both in the software and hardware, as
parking, assist diagnosing, assist investing, and assist piloting, etc. well as users of AI and it is dominated by Japanese and American
firms. The lack of many European stocks to play this theme may
So at the same time that collaborative robots are becoming more change in the future, but at the moment it is reflective, we think, of
economically viable for low-skill intensive, tedious jobs, AI is likely Europe falling behind in the innovation race, especially within the
to prove equally if not more disruptive to skilled labour. tech sector. Europes notable absence amongst the bigger
companies isnt due to a paucity of start-ups, but it has yet to
nurture them beyond adolescence. Further learning it seems is
Goldman Sachs Global Investment Research 5
Fortnightly Thoughts Issue 85

How to invest in this

Market Last Target Upside
Company Sector Country Rating Rationale
Cap($mn) Price Price Potential

NEC Japan Buy* 7,468 342 435 27% expressions)andtextmeaninganalysis(considerssentencestructureandimportanceof

Electronic ItsR&DResearchCenterremainstargetedontheIoTopportunity;activeinAdvancedDriver
Nidec Japan Buy* 18,696 8176 9000 10%
Components AssistanceSystems(ADAS)andautonomousdeliveryrobots

Mobileye Technology Nether. Neutral 8,657 $36.07 $50 39% anticipatingcollisionstowarndriver);ouranalystexpectsemiautonomousdrivingto
Nippon Electronic
Japan Buy 313 1578 2150 36% startofautonomousemergencybrakingassessments(from2016)andbroaderautosafety
Ceramic Components
Pacific Electronic
Japan Buy 449 1002 1170 17% alreadymandatoryintheUSandEurope,andgreater/mandatoryinstallationinJapanand
Industrial Components
Baidu Internet China Buy 73,334 $207.15 $241 16% 81%ofthetimeeveninnoisyenvironments;investmentsindeeplearning,language
Amazon Internet US Buy* 176,892 $373.37 $430 15% itsuniquesystemintegratesrecommendationsthroughoutthepurchasingprocessfrom
Twitter Internet US Buy 33,506 $47.82 $62 30% contentaroundworldevents(likeworldcup)betterremaintopproductprioritiesandfocus

WPP Media UK Buy* 28,683 1482p 1717p 16%

Criteo Internet France Buy 3,048 $43.43 $50 15% mediainhouse;highlyscaled,retargetingsoftwareplayerwithadifferentiatedmarket
Software US Buy 3,537 $57.52 $70 22% (voice,video,text)inhighvolumeandspeedusecases;leaderinActionableIntelligence
Opower Software US Buy 726 $13.65 $25 83% analysingover130Terabytesofutilitycustomerdata(eg:historicalusage,billing,weather)
Marketo Software US Buy 1,293 $27.02 $42 55% marketingautomationcapabilitiesaspermanagementoffersrealtimepersonalizationtech
Source: Datastream, Goldman Sachs Global Investment Research. *On the relevant regional Conviction List. Prices as of the close of February 17, 2015.

Goldman Sachs Global Investment Research 6

Fortnightly Thoughts Issue 85

Interview with Raj Rajkumar

Raj Rajkumar is a professor in the Department of Electrical and Computer Engineering and
Co-director of the General Motors-Carnegie Mellon Autonomous Driving Collaborative
Research Lab. He is the President and CEO of Ottomatika, Inc., was a primary Co-Founder or
TimeSys Corporation, and has served as a research staff member at IBM T.J. Watson
Research Labs.
When do you expect to see completely autonomous vehicles A wireless communication device which can be added to traffic
on roads? What are the key milestones between now and lights can transmit the location and current status of the traffic light
then? to a receiver implanted in a vehicle, allowing a self-driving car to
react appropriately. Similarly, this wireless system can also
I expect completely driverless cars to be feasible only in the 2020s,
facilitate vehicle-to-vehicle (V2V) and other vehicle-to-infrastructure
and even then it will first be offered in some developed countries,
(V2I) communications. For example, a car can communicate to the
where the infrastructure is more advanced and traffic rules are
traffic light about where it is going, so that the traffic light can adjust
generally obeyed. Autonomous vehicles will take longer to appear
its timing, turn green and let a car pass when there is no traffic in
in developing economies because: (a) their driving etiquette tends
other directions. This should significantly reduce delays and
to be very aggressive with very thin error margins, and (b) higher
pollution. Furthermore, such technologies will become available on
costs will initially be a deterrent. These issues will likely continue in
smartphones to enable vehicle-to-pedestrian (V2P)
the 2020s too. We can also expect that some rich, developed city-
communications, automatically alerting vehicles and pedestrians,
states or small countries like Singapore or Qatar, which are
when a pedestrian is crossing the street. It must be borne in mind
forward-looking in terms of adapting cutting-edge technologies, to
that the cost of upgrading traffic lights involves significant inertia.
become earlier adopters.
Fortunately, these communication technologies exist today, and
In terms of milestones, there are two very different ongoing
cars will be able to dynamically plan routes and co-ordinate with
paradigms. The first is the so-called Holy Grail approach
other vehicles. However, changes in infrastructure will only happen
exemplified currently by companies like Google, which has been
incrementally. We are beginning to see mandates in the US, in
experimenting on fully autonomous vehicles. Here the milestone is
Europe and in Japan to implement the wireless technology. Starting
really one capability - the vehicle will be able to drive to its
in about 2017 or 2018, the US government is expected to require
destination all by itself. But, another approach is the paradigm of
that all newly manufactured vehicles to carry this technology. But
incremental automation adopted by todays carmakers. Most if not
even so, given the whopping 250 million cars that can operate in
all car companies expect a human to be part of the driving process
the US and assuming that roughly 15 million new cars are sold
for quite some time to come. Even if the human is not necessarily
every year, it would take 16 to 17 years to replace the current fleet
driving, manufacturers expect a human driver to be present in the
of cars. So, it will take time.
drivers seat, engage and take over control from the self-driving
mode as and when necessary. That is still an advance from cruise What are the societal and economic benefits of self-driving
control, where right now, the driver gives up control over the vehicles?
accelerator pedal, but still operates the brake and steering wheel.
Every year an estimated 1.2 million people die and hundreds of
Most importantly, the driver has the responsibility of cancelling
millions of people get injured due to automotive accidents, taking a
cruise control and taking back full control when there is a vehicle
huge toll on society, both emotionally and economically. Imagine
driving at a lower speed ahead.
the potential benefits if we could take human error, which accounts
So, in this second paradigm, the first milestone - and its very for more than 90 percent of these accidents, out of the equation.
imminent - is an autopilot capability on the highway, allowing Unlike people who are prone to distraction, computers are not and
drivers to engage in a driving mode where the vehicle steers, they will be able to drive a lot safer. Another value-add of
brakes and accelerates in the current lane. The next development autonomous vehicles would be an improvement in productivity.
is a highway pilot capability with lane changes, where the vehicle Youd agree that being stuck in traffic is a very common urban
can also change lanes to pass a slow-moving vehicle in its lane. phenomenon. In the US, the normal person commutes about 51
Another milestone is a self-parking feature, with yet another being minutes per day to and from work, and thats unproductive time.
traffic jam assist (TJA) capability where the vehicle drives itself Over time, when the technology has proven itself, if the vehicle was
while in a traffic jam. When the traffic jam clears the human needs driving itself, that time could be used to catch up on email, texts, or
to take back control. Production cars available for purchase will doing something more productive including taking nap! I believe
reach these milestones in the next three to five years. My belief is that would have a global impact on economic productivity in time
that by the time vehicles become completely autonomous, people it is no longer a question of if, but when. Thirdly, people lose their
would hardly notice it because they would have slowly, but steadily, driving ability and license when they age. If older people are living
given up control for most of the driving functions before that final all by themselves, they would have lost some of their
step. independence, self-esteem and their quality of life would be
declining dramatically too. In this case, autonomous vehicles would
Does transportation infrastructure like roads and traffic lights
help them regain their mobility and could drastically improve their
need to change to facilitate the development of this
living standards. This is true for the disabled too.
What are the highest value-add components in an autonomous
It is very unlikely that infrastructure will change broadly enough to
accommodate this technology in the foreseeable future. As such,
autonomous vehicles are being built on the assumption that there An autonomous car comprises four sub-systems. The first one is a
will be no special infrastructure for driverless cars. That being said, set of sensors, things like cameras, laser scanners and radars.
infrastructural enhancements would be extremely beneficial for self- These act as the eyes and ears of the car. The second subsystem
driving cars. is actuators, which operate the steering wheel, the accelerator, the
Goldman Sachs Global Investment Research 7
Fortnightly Thoughts Issue 85

brake, and transmission; so, basically they act as our hands and technology is progressing rapidly, and the government wants to
feet. Thirdly, there is a set of computers that process all the facilitate this work, but ensure that it is safe. A few states in the US
information gathered by the sensors and then generate commands are also issuing their own rules to facilitate the testing and
to the actuators; in other words, this is the brain of the vehicle. And deployment of automated vehicles. In fact, these states are
fourth is a communication system that is similar to the human competing with each other to be perceived as the leader in this
nervous system; this receives sensory inputs from sensors and also technology by being very receptive to companies that want to build
communicates the commands from the brain to the actuators. and test self-driving vehicles. In return, they expect those
companies to set up businesses, proliferate their tech environment
Now, actuators are becoming cheaper and my view is that by about
and bring more jobs to their states. So competition has been a
2020, they will be broadly available and affordable. In terms of the
catalysing force and that is a very good thing.
sensors, cameras have already become very cheap and radars are
rapidly getting there, however lasers and scanners wont be as There are also very interesting developments occurring on the
affordable for consumers for a few more years. Essentially, the insurance front. As we previously mentioned, the rate of automotive
costs of a cars hardware components are subject to the same accidents, injuries and fatalities would go down with this
downward trend that is also evident in the computer and technology, which means that pay-outs for insurance claims would
smartphone industry. They are constantly becoming cheaper, faster drop. Subsequently, this would lead to lower premiums in a
and smaller. So the key to autonomous technology is really in the competitive landspace. So, of course, insurance companies have to
software which controls all these four subsystems. Driving is one of worry about the reliability of self-driving technologies, while being
the most complex activities that human beings engage in on a daily aware that the overall revenue and the market size could shrink
basis and so, any firms that become leaders in developing software significantly. If that happens, only the most nimble and forward-
for it will play a big role in deploying this technology, and that is looking companies are likely to survive or thrive.
where the bulk of the value will lie. And this situation is similar to
How do you see the use cases for wireless cars developing?
the Windows OS in PCs or Android in smartphones. There is also a
And what does that mean for aggregate car demand?
major role for someone who can vertically and smoothly integrate
the different function and capabilities of technology. So, looking Interestingly, Uber recently announced a strategic relationship with
again at smartphones, Apple obviously does this very well in terms us here at Carnegie Mellon University, to build self-driving cars for
of integrating hardware (design), software and brining together the car- and taxi-sharing market. And, almost immediately, Google
other related products, like the iPod and iPad. One can imagine responded saying it was looking into the ride-sharing market
something similar evolving in the autonomous car space as well. independently. The fact that two forward-looking giants are eyeing
this space can be an important catalyst to realise this driverless car
What happens when autonomous cars come across a new
technology, and I expect it to advance very rapidly. Of course, we
scenario on the road? Are they capable of adapting or have
dont know yet when exactly well start to see self-driving cars as
they begun to self-learn and improvise in new situations?
taxis, but I believe that other players will quickly join the game too,
In pretty much every project that is working towards autonomous and this on the whole will accelerate this technology.
cars around the world today, there is always a person in the drivers
A long-term impact of this sharing paradigm perhaps is in the
seat when the car is being tested on public roads, to act as the final
ownership model of cars. A common observation is that an average
authority and take over control if needed. The number of scenarios
car sits idle for more than 80% of the day, representing a big chunk
that one needs to test is very extensive. And multiply each
of dead investment. Instead, if there is a smaller subset of self-
individual scenario with different conditions in terms of weather,
driving cars which can simply be fetched or summoned by a
lighting, quality of roads, traffic, crowd densities and special
smartphone for a task, which, when finished, then drives away to
situations like a fire truck or an ambulance going by or a heavy
service the next person, then resources would be utilised much
truck right behind you. Driving really is a very complex activity for
more efficiently. This in fact would have a major impact on the
humans, but even more so for machines when they encounter
demand for cars, and could dramatically change the industry as we
unusual or unique scenarios. Someone has to teach these self-
know it now. Substantially fewer cars could be sold every year but
driving vehicles the right action to take, and in that sense they are
they may be replaced much more often due to the additional wear
not yet self-adapting or self-learning. Think about it this way - when
and tear incurred. Subsequently, fewer cars and better efficiency
we were young, we did learn by ourselves, but we were taught a lot
should have a meaningful impact on total resource and energy
more by our parents, our families and teachers. Eventually,
demand. And given a completely different user paradigm, car
because our brains are structured that way, we were able to learn
designs would evolve as well. More streamlined, efficient cars
things ourselves just by observing what was happening around us,
would contribute to greater energy savings. So, these are truly
based of the underlying principles that we were taught. My view is
exciting times for self-driving vehicle technology. Only a few years
that it will be a long time before self-driving vehicle technology can
back, people still perceived it as a long-term development i.e. that it
get to the position where it is able to teach itself what to do in
is 20-25 years away. Now, we are talking about what is going to be
different scenarios. We are still in the teaching phase.
available to consumers in the next five years. And that has been
What regulatory challenges to autonomous car adoption do because multiple milestones have been achieved in the past three
you foresee? years, and more companies are jumping into the fray. Remarkable
advances happened recently not just in terms of the technology, but
A lot of progress has been made on all these fronts. In the US,
also social awareness, social acceptance, insurance and the
regulations have been implemented at the central (federal) level;
legislative sides.
there is a very clear awareness at the federal level that this

Goldman Sachs Global Investment Research 8

Fortnightly Thoughts Issue 85

The road to automation

Advance of ADAS/connected cars to turn technology
Takafumi Hara, our Japan Electronic companies into auto-tech companies
Components analyst, on the future of ADAS penetration is starting to gain real momentum in the current
automated vehicles active safety era. Awareness of ADAS products is increasing and in
Europe and Japan the installation rate for automated braking is
high even for mass-market cars. With ACC and LKA also becoming
On the cusp of an automated driving era increasingly common in luxury cars, our impression is that the
The development of the automobile is nearing a major framework for semi-automated driving is gradually being put in
transformation point. The automated driving era is fast place. Amid the shift from active safety to semi-automated driving,
approaching, and the commercialization of related auto we expect the connected car penetration rate to increase rapidly
technologies is advancing at a rapid pace. We believe self-driving and the volume of tech products installed per vehicle (content value
vehicles will change lifestyles and create winners and losers in the per vehicle) to rise at a much higher speed.
technology sector (auto electronic parts). We forecast the market for ADAS sensing devices (millimeter wave
From a safety perspective, the development of automated driving radars, light detection and ranging (LIDAR) devices, cameras, and
can be divided into the following phases: (1) Passive safety: ultrasonic wave sensors used in automatic braking systems) will
airbags and anti-lock braking systems (ABS); (2) Active safety: grow quickly to US$10 bn in the next five years. This should be a
advanced driver assistance systems (ADAS), mainly automatic major tailwind for companies that make these products and the
braking, lane-keeping assistance (LKA), and automated parking; (3) electronic components used in them. We estimate auto electronic
Semi-automated driving: enhanced ADAS and connected car parts will become a pillar of tech company profits comparable with
technologies (vehicle-to-vehicle (V2V) and vehicle-to-infrastructure components for smartphones and other consumer electronics.
(V2I)); and (4) Automated driving: enhanced V2V/V2I. We are Companies with a high exposure to this field are likely to evolve into
currently in the active safety stage. ADAS is the core active safety auto tech companies.
technology and main ADAS functions include adaptive cruise
control (ACC), LKA, and automatic braking. These systems are The spread of connected cars should drive the market for ADAS
already found in vehicles and we expect the installation rate will sensors to US$10 bn
increase quickly. In addition to the initiatives of OEMs, non-auto Connected car global penetration rate and ADAS sensor market size
makers like Google and Japanese venture-company ZMP are ($US bn)
actively conducting self-drive related R&D. The roadmap to fully
12 70%
automated driving is filled with uncertainty and we do not expect
this stage of development to be reached until 2030 at the earliest. 60%
However, foundation technologies for semi-automated driving have 10

already been developed. We believe technologies like automatic 50%

braking, LKA, and ACC will make automated driving on highways a 8

reality by around 2020. 40%

ADAS and connected cars will open the door to the automated 30%

driving era 4

Automated driving development roadmap and changes in core 20%

1980~ 2000~ 2020~ 2030~

0 0%
Passive Safety Active Safety Semiautomated Fullautomated 2011 2012 2013 2015E 2020E
driving driving

Source: Goldman Sachs Global Investment Research, Japan Patent Office.

Airbag ADAS
Core ABS
ESC ACC LKA Drivermonitoring
AEB ParkingAssistance Night vision Linked The automation of driving and consumer pull demand are important
factors behind ADAS diffusion. However, we believe the most
ConnectedCars important driver of all is changes to safety regulations and safety
V2I (VehicletoInfrastructure)
assessments conducted by third parties. In Japan and Europe,
ABS:AntilockBrakingSystem automatic braking has been or will be added to the New Car
ACC:Adaptive CruiseControl
AEB:Autonomous EmergencyBraking Assessment Program (versus vehicles in 2014 and versus
LKA:LaneKeepingAssistance pedestrians in 2016). In the US, meanwhile, rear cameras will
become compulsory from 2018 under the Kids and Transportation
Source: Goldman Sachs Global Investment Research. Safety Act. Changes to the regulatory environment reflect rising
consumer safety awareness.

Goldman Sachs Global Investment Research 9

Fortnightly Thoughts Issue 85

high market shares in technologies that ADAS/connected cars are

Increasing importance of automated braking in vehicle safety likely to create new demand. As examples, we cite the following
assessments is promoting ADAS penetration companies in our coverage: Mobileye (UK), the No.1 maker of
NCAP roadmap in Japan, the US, and Europe mono-camera chips; Nippon Ceramic (Japan), the No.1 maker of
2014 2015 2016 2017 2018 2019 ultrasonic sensors for parking assistance; and Pacific Industrial
vs. Vehicle AEB vs. Pedestrian vs. Vehicle (Japan), the No. 2 maker of tire pressure monitoring systems
AEB (Path
(forward- AEB (Daytime)
interaction / (TPMS). While these companies only have small sales, they have
Right turn )
EU LDW already secured the top global market share in their product
vs. Pedestrian
AEB (Night

vs. Cyclists
AEB (Path Top specialists very attractive from the perspective of earning
growth rates
vs. Vehicle
AEB (forward- vs. Pedestrian
AEB (Daytime)
Comparison of 2013-16E operating profit CAGR: global auto parts and
Japan auto tech (JYP mn)


FCW vs. Vehicle AEB vs. Pedestrian 500,000

(forward- AEB (Night
LDW collision) time)
(US 400,000
vs. Pedestrian
NCAP) AEB (Daytime) Denso

LDP 300,000
Johnson Controls

Source: Denso, Goldman Sachs Global Investment Research. Magna

200,000 Delphi
Aisin Seiki

Winners from this trend: Unveil hidden gems 100,000
TRW Autoliv Lear BorgWarner
NGK Spark Plug
Stanley Electric Mobileye
We expect Tier 1 makers and automotive semiconductor and Toyota Boshoku
Keihin Calsonic Kansei
JAE (89%)
IRISO Electronics Nippon Ceramic
Unipres Nissin Kogyo
electronic component makers to be the main beneficiaries of rapid 0 Pacific Industrial
0% 5% 10% 15% 20% 25%
auto tech market expansion. For example, Tier 1 names Bosch,
Denso, and Continental; automotive semiconductor makers
Infineon, Renesas Electronics, and STMicroelectronics; and Source: Goldman Sachs Global Investment Research.
automotive electronic component makers Nidec, Murata Mfg., and
TE Connectivity. These companies have leading global market Automated driving technologies are increasing the IT product
shares in their respective fields. Compared with consumer content of automobiles. In other words, we expect the auto industry
electronics, the barriers to entry for auto tech products are high to converge with the tech industry in many fields. Rather than
because (1) it takes a long time to obtain spec verification and (2) bringing traditional auto companies and traditional tech companies
manufacturers must meet extremely stringent defective product into opposition, we think this trend will create new value for them
ratio standards. This bodes well in our view for large companies through reciprocal technology fusion.
that had already established automotive products as core
businesses before ADAS penetration.
But when considering investment in the auto tech industry, is it wise Takafumi Hara
to only invest in these top companies? As these companies tend to
Japan Electronic Components analyst
have broad product line-ups, we believe auto tech growth could be
diluted to some extent. From the perspective of earnings growth email: Goldman Sachs Japan Co., Ltd.
rates, we highlight first-class specialists that have already scaled Tel: 81-3-6437-9926
the high barriers to entry. In particular, we highlight companies with

Goldman Sachs Global Investment Research 10

Fortnightly Thoughts Issue 85

Interview withRodney Brooks

Rodney Brooks is the founder, chairman and CTO of Rethink Robotics. Previously, he co-
founded iRobot in 1990, where he served variously as CTO, Chairman and board member
until 2011. He also served as a Professor of Robotics at MIT from 1984 to 2010 and was the
founding Director of the Institutes Computer Science and Artificial Intelligence Laboratory. He
has been elected to the National Academy of Engineering and as a Fellow of the American
Academy of Arts and Sciences, the Association of Computing Machinery and the Association
for the Advancement of Artificial Intelligence, among others.
Baxter is an industrial robot built by Rethink Robotics and was introduced in September 2012.
It is a 3-foot tall, two-armed robot with an animated face that is designed to perform simple
tasks on a production line.
What are the changes occurring in the necessarily breakthrough, but they matter a lot to the improving
industrial robotics space and why do capability of robots.
you see an opportunity for
What kind of end-markets are you targeting for Baxter?
collaborative robots?
Currently, only about 10%, according to the Boston Consulting
Industrial robots are very adept in
Group, of manufacturing facilities and factories employ any sort of
situations that involve repetitive motions
industrial robots and they tend to be very high end; these are not
in a fixed co-ordinate system, but they
the areas that hold a lot of potential for collaborative robots. Baxter
havent evolved much in the past 50
is not going to compete in sectors where industrial robots make
years. Back then those robots didnt have sensors and computers
sense because they are already partitioned into distinct human and
connected to them, and today they still dont have built-in sensors
robot areas. Collaborative robots wont disrupt industrial robots
so they are not very safe to be around. Therefore, industrial robots
used in automotive manufacturing for instance. What we aim for
need to be put in cage-like enclosures to keep them away from
instead is to help put robots in places where automation was
people during the operating process.
previously not possible the incredible range of simple activities
Collaborative robots on the other hand use low-cost chips and that current robots are just not economical for.
sensors that enable them to impart a level of intelligence which they
Unlike traditional robots, which tend to be very expensive,
use for basic perception and reasoning. They are safe to be around
particularly when we consider the programming, installation and
and so can be employed in factories to work in close proximity with
integration costs, collaborative robots like Baxter are a lot cheaper
employees, instead of having a separate area dedicated to them.
to deploy. If we compare their cost and output against that of
Another important feature that distinguishes collaborative robots like
employing human labour in the US for instance, it takes less than a
Baxter is that ordinary people can use these robots and teach them
year to recoup investment. Further, these robots do not require
new, different tasks. They are extremely intuitive and flexible and
specialized engineers to integrate them for carrying out a single
integrate seamlessly with other automation technologies. They are
task and can be reprogrammed easily, which helps maximize ROI
also relatively affordable and do not require specialists to operate
as the same robot can be utilized across multiple lines in a plant.
them on a routine basis, and this is what makes them useful,
This is why these robots make an ideal fit for small manufacturers
particularly in smaller manufacturing facilities where processes and
with limited budgets, space and staff.
requirements change on a day to day basis.
What are the constraints that have hindered greater adoption
What is driving the improvement in economics for
of collaborative robots so far?
collaborative robots?
Collaborative robots are still quite new as a category and the lack of
Improvements in capability and a persistent reduction in the cost of
understanding of what they do, how to use them and how they
hardware technologies, thanks to the smartphone revolution, have
differ from other industrial robots has been one of the key
driven a lot of progress in the field of collaborative robots. Rapid
bottlenecks that we have faced. It took us a while to get people to
growth for electronics in the last 20 years has helped bring down
understand how these robots work and why mapping their
the cost of cameras, sensors, accelerometers and other related
capabilities with existing industrial robots doesnt work. But
technologies, and this has enabled the cost of collaborative robots
adoption is now picking up.
to come down, because these robots use processors and sensors
that are the same as the ones embedded in smartphones and other We have also improved the software inside our robots dramatically
electronic goods. in recent months. For Baxter for instance, upgrades from last July
have made the robot even more capable and productive than it
The other important development of course has been the increasing
used to be. It is able to do the kind of tasks that people want it to do
role of software in robotics; software reduces the time and cost of
more and more and that should help boost penetration further. We
installing, maintaining and modifying robots. It also makes the robot
are continually expanding the capabilities of the robot and I think
more flexible, enabling it to impart an intuitive user interface and
we will see demand get way out in front of our supply this year as
helps the robot improve over time, thanks to regular updates.
we ramp up production.
The latest upgrade of the software that powers Baxter, for instance,
helps users to re-deploy the robot quicker and more easily, after
common plant-floor variations occur, such as tables being bumped,
fixtures being moved etc. These individual upgrades arent

Goldman Sachs Global Investment Research 11

Fortnightly Thoughts Issue 85

How do you see software usage in collaborative robots You've refuted the concerns around the potential dangers of
evolving? artificial intelligence and negative impact on jobs...
It is software that adds new capabilities to the robot. And so, with Yes. Things like deep learning have made significant progress in
Baxter we are aiming to merge our software programs into one the last five years; very surprising progress, and this has made
package and make it open to third-party developers, at every level, many outsiders to the field overestimate the capabilities of the
in 2016. This will allow lots of people to build new capabilities for computer programs that are now able to do some of the things that
the robot, which could then be sold by third parties or remain in an human beings do. But while programs can now label images based
open source format. on their content, for instance, they do not understand what the
Hundreds of people are already developing software for Baxter in image really means. So, if a collection of randomly organized
universities. And while all of the new developments may not be pictures of a dolls parts are fed into a computer program, it might
useful to our broader customer base, there will surely be some just label the image as a doll without understanding the concept of
good ideas that will emerge as so many different smart people work a doll. A person, on the other hand, will be able to say that the
on improving the capabilities of the robot. image is not that of a doll, but a collection of doll parts. So I believe
people outside the field over-generalize the capabilities of what turn
What is the competitive intensity like in this category of out to be very commercially useful programs, but where the
robots? adjacencies don't generalize in the same way they would if a
While collaborative robots did not exist as a category until as person was doing that task. It is extremely hard to assert what the
recently as a year and a half ago, the big four robot manufacturers, capabilities of artificial intelligence will be like 100 years from now,
i.e. Fanuc and Yaskawa from Japan and Kuka and ABB from but from a 30- to 50-year perspective, I do believe there is a gross
Europe, and smaller companies like Universal Robots, have all overestimation of the capabilities of artificial intelligence which is
already offered or are in the process of developing some sort of leading to worries and doomsday scenarios. Also, while there will be
collaborative robot or the other. Some of them are making some disruptions in the labor markets as artificial intelligence
advances, but currently they are at a much higher cost compared to becomes more capable, that has always been the case with
Baxter. We also have a lot of patent protection, which will make it technological progress. Overall, the progress being made in this
difficult for other players to do exactly what we're doing. But we are field is likely to help people do their jobs better rather than replacing
glad that big companies are foraying into this domain. It not only them completely.
validates us and what we do, but also opens up more possibilities How accommodative are regulations for collaborative robots?
for the field of collaborative robots.
The current industrial robots are not safe to be deployed in
In which segments and geographies are your customers proximity to humans and the safety standards established for them
concentrated? do a good job at maintaining distance. But a robot like Baxter does
Our customers come from a variety of sectors, from small plastics not need to be separated from people on the factory floor. And this
manufacturers to third-party logistics. Areas like machine tending is what we are trying to explain to all the standard setting bodies in
(putting sheet metal in press brakes), which is both boring and this field. It has been easier to do that in some geographies and
dangerous in nature, also represents a growing market for us and harder in others.
we believe robots can be perfect substitutes for risky and dull jobs There are also many incumbents, especially in the safety industry,
like these that people are not keen on doing. which are trying to invoke regulation on safety grounds, but again,
Apart from this, I believe that elderly care will be one of the biggest this isnt the first time that innovation is faced with regulatory
drivers of robotics over the next 30 years. The demographic hurdles. Historically, train companies tried to use regulations to stop
inversion in Japan, North America, Europe, Australia and even people from using automobiles in Great Britain in the 19th century,
China simply necessitates greater adoption of collaborative robots. and we are seeing similar oppositions being hurled at Uber
Now, while Baxter has some capabilities that the previous robots currently. So, like many other innovations, there will be a process
did not have, it is still not in the perfect form needed for elderly involved and it will take some time to straighten regulations for
care. And this is why it is great that Baxter is being used in collaborative robots, but eventually, I dont think regulation will
universities and that a lot of people are trying to work towards stand against progress being made in the field.
developing elder care applications with Baxter.
How do you feel about the prospects of innovation more
In terms of geography, we are mostly located in the US and broadly?
Canada currently, but we plan to expand internationally later this Some of the biggest problems in the world require a little longer-
year. But production should still be limited to the US, where we term development of not just the software but also the hardware
design and build Baxter today simply because while designing the coupled with it. And so, while a lot of software-based platforms
robot, we worked alongside 19 different custom parts suppliers to have seen valuations rocket and in turn led to a situation where
figure out where their low-cost sweet spot was to optimize Baxters many VCs are focused on shorter-term hits, we need longer-term
production, and fed that information into the design process. And investments to come up with genuine innovations that can help
they are all in the US and Canada. I dont think we could have done solve the worlds constraints.
that in China without going through an extremely long and
expensive process and even if we did, I dont think we could have
built our robots as cheaply in China as we build them here in the US.

Goldman Sachs Global Investment Research 12

Fortnightly Thoughts Issue 85

The start of the collaboration era

Machinery manufacturers are seeking to capture various growth
We asked Yuichiro Isayama, our Japan opportunities in an effort to alter this trend. A good example is the
Nextage robot produced by Kawada Technologies. This robot is
Machinery analyst, how collaborative robots able to directly replace a human operator for light and small tasks.
are changing the Machinery industry Equipped with vision sensors in the head and arm sections, the
Nextage robot can complete simple, light and small assembly
Japan's machinery industry is home to many leading suppliers of processes instead of a human operator. With Nextage retailing at
machinery and equipment that fall under the category of Factory just over 10 mn including system costs, the price has come down
Automation (Discrete Automation). Through a process of trial and to a level that allows potential users to give serious consideration to
error, Japanese machinery makers have been pioneering the replacing human tasks with robots. Based on announcements, it
development of robotics/the automation industry. The majority of seems that both Japanese and foreign companies such as Glory
companies in Japans machinery sector are focused on cementing and Airbus have introduced Nextage, which is steadily building up a
their position as suppliers, rather than actual makers of core track record already.
technologies of artificial intelligence or controlling systems of those Kawada Technologies assembles Nextage robots and developed
smart machines. its control system, while THK supplies the components. THK also
The robot is probably the best example of a machine that features provides the required teaching and integration services when a
both collaborative and learning intelligence functions. In recent client chooses to install Nextage. We believe this offers major clues
years, Fanuc, the top global industrial robot manufacturer, has as to how the leading automation players in Japan's machinery
unveiled a prototype collaborative robot that has attracted industry plan to approach the AI and Smart Machine space.
considerable attention, while also much talked about is the Nextage THK's CEO Akihiro Teramachi explained in an analyst meeting last
collaborative robot, co-developed by automated machine September that he saw little value-added in assembly per se. This
manufacturer THK and Kawada Technologies, a supplier of seems reasonable to us given that automation-related equipment
helicopters and equipment for defense applications. Below, we use and component suppliers like THK maintain higher margins than
these two examples to discuss the features, current themes, and manufacturers involved only in robots and other assembly
outlook for collaborative robots. processes. Also having a major impact, in our opinion, is that the
Fanuc produced and demoed a prototype collaborative robot in standardization/establishment of control systems is the most
2014 following the relaxation of ISO regulations (allowing the difficult area for Japanese companies to rival their US/European
removal of safety fencing surrounding robots with a power output of counterparts.
less than 80kW). Ensuring the safety of peripheral workers is the
biggest precaution in human/robot collaborative processes. Many Global operational stock of robot by industry
touch and vision sensors usually need to be fitted to prevent any Operational stock of industrial robots globally
accidental contact between human and robot. With its collaborative 500,000
system, however, Fanuc has slowed the movement of the vertically 450,000

articulated robot to a fraction of the regular speed and covered the 400,000

robot with a green sponge and installed a sensor to shut down the 350,000

robot's operation upon human contact. By doing so, Fanuc has 300,000

minimized the actual impact on the human operator from contact 250,000

with the robot. The only additions to a standard vertically articulated 200,000

robot are the sponge, the sensor and the reduction in speed. Since
the function of the robot is little changed, Fanuc has successfully
kept down its price vs. complicated collaborative robots.


Plastic and chemicals


Foods and beverages

Other mfg. branches

It seems potential users of Fanuc's collaborative robot are mainly

automotive and electronics clients that already employ vertically
articulated robots in significant numbers. Nearly all robots used in
manufacturing at present are not self-organizing, but instead
require humans to teach them process patterns. For a company
considering introducing industrial robots in general, the key Source: Goldman Sachs Global Investment Research.
determinant for opting to automate a process is whether or not it
has sufficient experience in using robots already. We believe this is
why the main potential users of collaborative robots are yet
concentrated in the auto and electronics industries, and why we still
see considerable room for growth in robot penetration chiefly in
general industry, as shown in the next exhibit.

Goldman Sachs Global Investment Research 13

Fortnightly Thoughts Issue 85

Our interviews with companies operating in new automation/robot Economy, Trade and Industry and other parties that there is no
fields indicate to us that Japanese manufacturers' chances of growing momentum to create specifications unique to Japan.
success are low in the both the self-organizing space, typified by
We believe a similar situation is occurring in the self-organizing
AI, and the control function space, as these areas require the sort
space, and this is why most Japanese machinery manufacturers
of aggressive strategy with which they have generally struggled
have not stepped up their own development of software but prefer
historically. First, both spaces are software based, and international
to play a role as equipment/component suppliers. In the new era of
standardization is unfamiliar territory for Japanese manufacturers.
robots and automation, Japan's machinery sector looks likely to
Second, fundamental to Japanese manufacturing industry as a
retain its position as key supplier.
whole is an ethos of offsetting deficiencies in software functions
with hardware. Third, Japanese companies do not have the
financial wherewithal to rival major US/European IT players where Yuichiro Isayama
large-scale investment is required from the development stage. Japan Machinery analyst
Although Japan sits on specification committees for IoT email: Goldman Sachs Japan Co.,
standardization (Industrie 4.0 in Germany, Industrial Internet in the
Tel: 81-3-6437-9806
US, etc.), we confirmed via interviews with Japan's Ministry of

Goldman Sachs Global Investment Research 14

Fortnightly Thoughts Issue 85

Interview withManoj Saxena

Manoj Saxena is a founding General Partner of The Entrepreneurs Fund, where he focuses
on identifying, investing in, and scaling cloud-centric software companies in the Cognitive
Computing space that also complement the IBM Watson Cloud ecosystem. He is a Special
Advisor to IBM senior leadership, where he focuses on operationalizing IBMs $100 mn
Watson Cloud Ecosystem Fund and making side-by-side investments with the TEF IV fund.
Prior to joining TEF, Saxena was General Manager, IBM Watson, where his team built the
worlds first cognitive systems in Healthcare, Financial Services and Retail. Prior to IBM, Mr.
Saxena successfully founded, scaled, and sold two venture-backed software companies within
a five-year span.
How have cognitive computing and artificial intelligence people understand and embrace cognitive systems, as well as the
evolved in the last few decades? And how do you see them development of new tools and methods. Inevitably, IT and business
developing in the future? people alike will start looking at the new paradigm for creating
value, while recognizing cognitive computing not just as a way to
Cognitive computing, as an incredibly useful and transformative
accelerate a companys investments in Big Data, but as a
technology, has reached the tipping point, in my opinion. The
fundamentally new platform that will provide exciting opportunities.
technologies that enable cognitive computing at scale have been
around for at least one, if not two or more decades. And right now, Which industries have begun adopting cognitive computing
these multiple technologies are converging in a perfect storm, and where else do you see potential in the future?
creating tremendous opportunities for businesses.
Cognitive computing has the potential to become omnipresent in
Firstly, cloud technology has meant that massive computing capacity every industry that we know of, but there are some industries that
and near infinite processing power is now available at very low prices are, naturally, adopting this technology sooner than others. These
and requires no upfront installation or capital spend. Secondly, new are information-intensive industries, where access to data is
machine learning technology and its applications to different functions, creating a tremendous number of opportunities. This includes
like natural language processing, are maturing. Take Apples Siri, industries like healthcare, insurance, telecoms, financial services,
Nuance, IBMs Watson, Microsofts Cortana and Googles efforts in and travel. Within these industries, there are the early movers that
this direction, for instance. Thirdly, there has been rapid proliferation are already adopting cognitive computing. And to my surprise, this
of mobile devices and the emergence of a new class of users, which shift is happening faster than I had anticipated.
together are creating a new distribution mechanism. And last but not
Were also seeing cognitive clouds being adopted across
least, we are also seeing substantial amounts of data being
organizations. Specifically, across horizontal business processes
generated from both social devices and the Internet of Things, in a
such as procurement, customer care, HR, sales, manufacturing, and
form that traditional computers cannot process and extract
marketing. For these functions, cognitive insights enable companies
intelligence from. This data is known as dark data. It contains critical
to better understand the thousands, if not millions, of business
insights and makes up more than 80% of data today, yet it often goes
processes transacted throughout their companies every day. This
uncollected or neglected. One of our portfolio companies, Cognitive
insight transforms business processes, which boosts productivity,
Scale, is able to extract insights from dark data to help customers
improves customer satisfaction, and results in cost savings.
improve decision-making, personalize consumer experiences, and
create more profitable relationships. Across sectors, there are mainly three selling points for AI or
cognitive computing a greater ability to provide highly
Also, we now have an entirely new set of workers and consumers,
personalised insights and experiences to customers, enabling
millennials, who are starting to use these technologies, which are
better decision-making in a timely manner, and lastly, creating more
converging to enable the endless capabilities that cognitive
profitable relationships with customers and suppliers. Cognitive
computing represents. I believe, along with IT industry analysts at
computing is going to cause a big shift in enterprise and consumer
IDC, that cognitive computing is creating the third big platform in IT. It
applications; in other words, every application will have analytics
is following the first two, which are marked by mainframes and personal
woven into it. So, instead of just laying out the processes without
computers and followed by the World Wide Web. There will be very
taking into account any context, in the future these applications will
exciting opportunities in the next few years, as the cognitive
start integrating intelligence from third-party data. For example,
computing platform leverages the convergence of cloud, mobile and
when you are using an application to process someone's mortgage,
big data and I expect this trend to last for the next couple of decades.
with the cognitive platform you can go beyond looking at the
The rate at which cognitive computing can accelerate in the future information provided to you on the form and very easily look at the
will depend on three major factors. Firstly, platforms will need to applicants social profile, the kinds of activities that they are
implement these technologies and deliver results more quickly, involved in or the likelihood of flooding in the area of the house
perhaps in a matter of weeks, rather than the current period of nine being underwritten for mortgage, or even claims on properties
months to a year. So basically, there should be specialized platforms getting vandalised from police reports. Such third-party data
to create and process cognitive applications speedily. Secondly, as facilitates smarter business processes because the analytics-
businesses start to see initial pilots pay off, they will invest heavily woven applications are able to understand unstructured data, as
in many more projects, just like they did with the internet. The well as text and images altogether.
moment these businesses recognize the breakthrough value of
This point can be illustrated by an example of Cognitive Scale
cognitive computing, they will implement this technology in all of
application in a children's hospital. In large cities, care providers are
their applications and processes, a phenomenon that we have
looking after tens of thousands of kids for diseases like paediatric
begun to see today. And lastly, there will be an evolution of how
asthma and the key aim is to prevent the condition from getting

Goldman Sachs Global Investment Research 15

Fortnightly Thoughts Issue 85

worse for those afflicted with it. Currently, this is done by going third-party and fourth-party data. Then there is also a barrier from
through different touch points: from personal records to lab reports the lack of skilled labour to build out these applications and the
and then following up with parents and the kids to get them treated. need for a new governance model to embrace this technology. Last
All these, however, only take physical data into account, without but not least, the markets have to learn and understand the myriad
considering lifestyle, socio-economic factors, weather data and so possibilities that this technology can bring. These things are
on. Cognitive Scale can overcome this problem by connecting the creating obstacles to adoption, but are currently being tackled by
process to third-party data like US census,, various players., Google Maps, and ZocDoc which shows which
Who do you think will benefit from this technology and who
doctors are available and how highly they are rated. So for
will be the losers?
instance, when predicts a ragweed outbreak in a
particular zip code, the cognitive application can quickly identify the Cognitive computing is definitely going to create a massive shift in
number of children in that area who have an allergy to ragweed. productivity, efficiency and innovation. Productivity wise, hundreds
Then based on the US census map, the application can start of these systems are going to raise the proficiency of every worker
dividing the people into those who have socio-economic issues and in different businesses to the level of the best knowledge worker in
where access to transportation is a problem. Finally, it can suggest his or her field, and beyond. In essence, they allow democratization
a few solutions, such as getting the children to a nearer clinic, of best practices, thus enhancing the efficiency and capability of
shipping them an inhaler or calling the nurses at their school to each worker. Furthermore, this should also foster the emergence of
keep the kids inside, away from the ragweed outbreak. This new innovative business models. Just understanding the outcome
example demonstrated how intelligence from third-party data can of the reward-sharing models could lead to the creation of new,
be woven into a business process, which traditionally was not different and customized customer offers, for instance. Take the
possible, to guide a knowledge worker in making better decisions mortgage example from earlier; a traditional procedure to process a
and to improve relationships with their patients. Cognitive systems mortgage would be more labor-based and involve document
are going to transform every business process that we know and translation, application checking and filling and so on. If cognitive
create myriad opportunities. intelligence is woven in, not only would the application process
mortgages, but it would also help the customer reduce liquidity
Can we expect one ultimate dominant cognitive system or is
requirements, thanks to a better understanding of risk.
the market still fairly open? What are the barriers to entry for
cognitive computing? In that context, providers of old technology will lose out if they are
unable to come up with new innovative solutions fast enough.
It is still too early to tell as we are probably in the third year of a
Without intelligence woven into their solutions, these last-
seven-to-ten year cycle of growth and maturity. If we refer to the
generation technologies are at massive risk of getting
lessons we learnt from the early days of the previous two tech
disintermediated by a new class of vendors like Cognitive Scale or
platforms, we can expect dozens of new companies to come about
Spark Cognition, just like the shift to cloud has disrupted revenue
and change the industry in the following four dimensions: At first,
streams and pricing models for many giant tech companies, thus
there will be many point-to-solution vendors who provide just deep
reshaping the entire landscape. In general, cognitive computing can
learning algorithms; many of these are going to struggle and some
be disruptive, not just for software vendors but for software
of them will fall by the wayside. Then we will see the emergence of
infrastructure vendors as well. In addition, companies that are slow
large companies bringing out platforms to start creating cognitive
to adopt cognitive computing will lose their competitive advantage
applications. Subsequently, specialized vendors will start providing
because their competitors can improve the experience of their end
platforms and applications that are optimized to specific industries,
users, make better decisions and improve profitable relationships.
giving inch-wide and mile-deep solutions to really address hard-core
business problems. And finally, system integrators will start providing And if cognitive computing can help knowledge workers make
consulting and deployment services around cognitive clouds and better decisions, when do you think they will replace workers
cognitive solutions. altogether?
Cognitive Scale is a very interesting case in this context, because I think that we are still very far away from that, at least 20 or 30
large platforms often fail to provide highly focused and pinpointed years. The concerns about job losses have been somewhat
solutions that deliver quick value. This is perhaps due to three overdone, as has been the case with every shift in technology.
problems with the cognitive systems it is expensive to put Consider the forecasts versus reality when the telegraph came out,
everything together as a technology stack, once they put it together then the telephone and the internet. What is more important is how
they need to deal with the fact that the technology is still changing, cognitive computing will augment human cognition and decision-
and last, but not least, there is insufficient skilled labour to build a making, rather than replacing human beings. In other words, it will
solution on top of it. As a result, we will see the emergence of many be more about man and machine, not man versus machine. In the
smaller competitors who will address narrow and core problems. As end, what is probably not known is the scale of disruption and the
these new competitors create their own niches, they will expand opportunities brought along with cognitive computing. We are
fairly quickly if the big companies don't respond. The industry saw entering an era that mirrors the advent of the internet in the mid-to-
this when Documentum created a document-centred database and late 1990s, when people moved from client server-based systems,
when Logic created a platform for internet application. Overall, I and the markets grew exponentially with companies worth billions
think this space is just getting going and it's not anywhere near being created, and new vendors and new leaders emerging. The
overcrowded. same thing is going to happen here, and it is occurring even faster
than I had anticipated. My advice would be to keep an eye on the
Of course, we should keep in mind the major constraints in
smaller players, because they are more agile and innovative, and
adoption of cognitive systems. Beside the availability of platforms
they are learning at the cost of big players, which are still opening
that can create rapid applications, the availability of data to feed
up the school bus.
into these systems also plays an important role. This role includes

Goldman Sachs Global Investment Research 16

Fortnightly Thoughts Issue 85

AI is this time different?

Machine Learning: Generally refers to algorithms that run on large
Our software analyst Greg Dunham stresses data sets and create a model of inputs that can then be used to
predict or suggest outcomes given a set of variables. These models
that Big Data technologies can bring in a new are dynamic, and change as new data enters the data set, as
set of Artificial Intelligence Applications opposed to traditional static programming instructions.
Optical Character Recognition: Generally refers to software that
While the emergence of Artificial Intelligence (i.e., thinking analyzes printed documents or scanned images of documents, and
machines) dates back decades, new cloud models in conjunction generates machine-encoded text from the source material. This
with the advancements in processing and big data technologies technology simulates the human ability to read and internalize
have enabled use cases to expand. Tech industry heavyweights printed text.
such as Google (Google Now), Microsoft (Cortana), and Apple (Siri)
are all investing aggressively in the areas of Machine Learning and Semantic Network/Semantic Net: A virtual structure that links
Natural Language Processing to bring to life intelligent virtual concepts with related concepts, and words with related words.
assistants which leverage the broadening set of data (voice, Semantic nets are often used in natural language processing in
location, web, etc.) with personal history to understand how to order to allow algorithms to determine context for queries and
serve the consumer best. improve responses.

Awareness has also grown as speech recognition, from companies Hadoop: An open source framework that uses low-cost commodity
like Nuance, is increasingly embedded in consumer electronics hardware to process and store large data sets. Core Hadoop
(i.e., smartTVs) and cars, and consumers have now seen how a consists of several open source components including Hadoop
computer (i.e., IBMs Watson) can win the TV gameshow Jeopardy! Distributed File System (HDFS), YARN, and MapReduce.

Perhaps more interesting is the emergence of hundreds of start-ups Mahout: Open source, machine learning library of algorithms
which are leveraging new open source technologies (i.e., Hadoop, designed to run on very large, scalable, distributed systems,
Mahout, Spark), often with cloud models, to create a broad array of especially Hadoop.
recommendation engines, enable programmatic bidding within
Watson: Computer system developed by IBM that uses machine
advertising, reduce fraud, optimize a manufacturing process, and
learning and natural language processing to generate answers to
even help detect prostate cancer.
common-language queries. Famously, Watson competed on the TV
Start-ups like Scaled Inference and Sentient have raised tens of gameshow Jeopardy! against previous champions and consistently
millions of dollars in VC funding to better bring to market solutions outperformed its human opponents.
which deliver real business value, something which has historically
Virtual assistants: Software programs that use natural language
been elusive in AI business models.
processing, and machine learning to simulate an actual person
when interacting with humans. Typical applications include
Percentage of Indeed job postings by keyword since 2005 automated customer service, and voice-activated modules such as
Apples Siri. Also referred to as automated assistants, automated
online assistants, or digital personal assistants.
Spark: Open source framework designed to perform high-speed,
repetitive queries on data stored in-memory. Highly scalable
technology, as it is leverages cluster computing. Spark can function
with several other open source frameworks, including Hadoop
(HDFS), Cassandra, and others.

Private companies, and public investments in artificial intelligence

Investments by tech companies
Google acquires DNNresearch and DeepMind
Yahoo! Acquires IQ Engines, LookFlow, and SkyPhrase
Pinterest acquires Visual Graph
Dropbox acquires Anchovi Labs
Source: Facebook acquires
Digital Personal Assitants from Google (Now), Apple (Siri), and Microsoft
Private companies
Artificial Intelligence (AI): Generally refers to software that can Sentient Technologies Vicarious
interact with humans in a way that makes it appear as if the Scaled Inference Narrative Science
software has a human-like ability to reason. Two major components Expect Labs Kensho
of Artificial Intelligence are Natural Language Processing and MetaMind Nara Logics
Tempo AI
Machine Learning.
Source: gigaom, geekwire, Goldman Sachs Global Investment Research
Natural Language Processing (NLP): Analysis of interactions
between human language (natural language) and computers or
other machines. The goal of natural language processing is to allow Greg Dunham, CFA
computers to fully understand normal spoken and written language,
without the need to adjust that language to fit computer or software US Software analyst
language rules. email: Goldman, Sachs & Co.
Tel: 1-212-357-7217

Goldman Sachs Global Investment Research 17

Fortnightly Thoughts Issue 85

The age of robo advisors

What is automated investing?
Alex Blostein, our US Capital Markets & Automated investing (dubbed by the press as robo advisors)
Exchanges analyst, discusses the potential offers a low-cost, 100% web-based portfolio allocation solution to
the client by using software to make investment and rebalancing
and challenges for automated investing decisions. The platform creates a portfolio based on a clients age,
risk preference, and income that consists entirely of low-cost ETFs
Automated Investing Solutions a fad or a real game changer across asset classes (US Stocks, Foreign Stocks, Emerging
in wealth management? Market, etc.). The portfolio is automatically rebalanced using algos,
keeping behavioral bias aside. These platforms also provide other
Over the last two decades, innovation in Financial Services led to a
services such as tax loss harvesting and single stock diversification.
wide range of benefits, including greater transparency, lower cost
and improved speed of transacting business. For retail investors, Since their launch in 2010-11, automated investing firms have
the frontier has largely focused on two elements lowering cost garnered over $15 bn in AuM in the US still a rounding error
and enabling self-directed investing, with the use of technology. compared to the $12 tn of total assets under management. Some of
The origins of this trend date back to the mid-1990s, when the larger players in the space include Betterment and Wealthfront,
electronic brokers such as E*Trade, Ameritrade, Charles Schwab, but Schwab has also recently announced the launch of automated
and Fidelity made trading more convenient and less costly by advisory solutions (Intelligent Portfolios). The former charge about
displacing voice brokers and moving the process online. Coupled 25 bp on a clients AUM as an all-in fee, while Schwab is providing
with evolution of the asset management product and growth of the service for free.
passive investing (index funds and ETFs now comprise over 30%
of equity assets in the US), end-user fees continue to decline and The average account size at these firms is notably smaller than
switching costs become lower. traditional wealth managers, averaging ~$90K. Moreover, clients
are significantly skewed toward younger generations, with 90% of
The next area where we might see innovative disruption appears to account holders being under 50, and 50% under 35.
be in automated investing (robo advisors) a form of algorithmic,
software-driven financial advice that cuts human advisors out of the
equation. By improving the scalability of advice with technology,
US household demographic
and thereby significantly reducing the cost to end clients,
$200,000 50%
automated investing can engage a whole new segment of the
population that traditional managers have never touched: largely
younger, small account-holders. $160,000 40%

While we think that in the near term, the potential disruption will be
limited as automated advisors gather assets mainly from clients $120,000 30%
that are too small for a traditional model, longer term, automated
investing has the potential to gain share as the transfer of wealth
from baby boomers to the next generation puts tech-minded $80,000 20%

Millennials in the drivers seat. That said, the robo advisor solution
is not without its challenges, as these models are yet to be tested
$40,000 10%
by a market downturn, barriers to entry are low, scale of larger
competitors could keep profitability for new participants under
pressure, and the missing human element could be too difficult to $0 0%
overcome as someones wealth grows. < 35 35 - 44 45 - 54 55 - 64 > 65
% of US Pop. (RHS) Avg Net Worth (LHS)

Source: US Bureau of Labor; 2010 Census; Goldman Sachs Global Investment

Fee rates of different types of brokers Research.

Merits of the offering:
Advice at a lower cost: Technologically enabled low cost to
end-user is a clear advantage of the automated investing offering.
0.7% Without the traditional wealth managers brick and mortar offices,
0.6% the all-in fee of just 25 bp (and free at Schwab) is on average 50
bp lower than the fee rate that incumbents charge. More
importantly, the 25 bp fee rate is also 30 bp lower than the cost
0.4% basis (expenses/client assets) at incumbent firms.
0.3% The Millennial factor: Millennials (born between the 1980s and
0.2% 2000s) are the primary client targets of the robo advisory
industry. This group tends to be more conservative and more
focused on achieving personal finance goals vs. beating the
SMID Brokers Wirehouses eBrokers Automated invest. rev rate
market. Moreover, a Cerulli survey suggests that a greater
Revenue Rate (ann.) Expense Rate (ann.) proportion of Millennials value online advice (27% vs. 12%) and
Note: "Wirehouses" include: BAC, MS, UBS, WFC; "SMID Brokers" include RJF, goal planning (20% vs. 10%) than older generations. At an
SF, LPLA, AMP, "eBrokers" include: ETFC, SCHW, AMTD estimated 80 mn (or ~25% of the US population), the group is
Source: Goldman Sachs Global Investment Research.
bound to eventually control a meaningful portion of wealth.
While currently, nearly 70% of the total net worth in the US is
held by the baby boomers (age 55 or older, 25% of the US
population), according to Accenture over $30 tn in wealth is

Goldman Sachs Global Investment Research 18

Fortnightly Thoughts Issue 85

expected to be transferred from the Boomers to younger Limited track record: Because automated investment firms
generations. Thus, if automated investment advisors are were established in 2010-11, and we have not experienced a
successful in retaining their younger clientele today, there could full economic cycle since then, there is no historical record on
be a meaningful threat to the traditional model 20-30 years from how the algorithms used by automated investment firms will
now. react under drastic market conditions. Thus, it may be difficult
A sweet spot of engaging the under the radar assets: for automated investment firms to convince potential clients that
Today, only 30% of those under 30 in the US receive regular the returns on their portfolios can hold up in a down market.
financial advice. However, about 2/3 of that population said they The trust factor: Despite the cost benefits of using
need more financial advice and 81% said they want to be more technology, the generations that carry the most wealth (i.e.
involved in the daily management of their portfolios, per Cerulli. baby boomers) today tend to prefer human advisors over
This segment represents an attractive opportunity for software and algorithms. Thus, the ability for automated wealth
automated investment firms to go after given its low-cost advisors to gain share from traditional wealth mangers and
approach, as this group is unprofitable for most traditional tech-assisted advisors might be constrained in the near term.
models where costs are higher. Moreover, traditional wealth advisors often offer a more holistic
spectrum of services throughout the clients life cycle (Trust and
Potential challenges: Estate planning, financial planning around specific liquidity needs
Low barriers to entry: We think the barriers to entry in the and life events, etc.), that algorithms may not be able to perform.
automated advice space are low. To that extent, we have
already seen similar solutions being developed in-house, such
as Charles Schwabs Intelligent Portfolios (provided free of Alexander Blostein, CFA
charge). While there are nuanced differences between each US Capital Markets & Exchanges analyst
offering, the core of the product is fairly similar automatic
email: Goldman, Sachs & Co.
rebalancing, tax efficacies, etc. and is already being offered by Tel: 1-212-357-9976
others. Moreover, a Cerulli survey shows that beyond the 30%
of traditional wealth advisors who have already announced
plans to launch an automated investment management tool,
about 64% have indicated interest in developing such tools.

Goldman Sachs Global Investment Research 19

Fortnightly Thoughts Issue 85

Six of the best our favourite charts

In our six of the best section, we pull together a pot pourri of charts that we hope you will find
interesting. They will be different in each edition but hopefully always of note.

Summer 2014 things started getting cheaper Energy demand, blowing away
Producer Price Index for selected countries, Jan 2010 = 100 Danish gross energy consumption, 1972-2014 (TJ fuel equivalent)
160 900,000

150 Russia 880,000

Brazil 860,000
130 India
UK 780,000

100 S.Korea 760,000

90 740,000











































Source: Datastream Source: Danish Energy Agency

Decade highs, not all-time highs Russian twists

Aggregate funding to US start-ups ($ billions) and number of deals YTD performance in USD, as of February 16, 2015
investment deals (RHS) Russia (MICEX)
$ 30 3000 S.Arabia (Tadawul all)
Sweden (OMX Stockholm 30)
Italy (FTSE MIB)
Argentina (Merval)
$ 25 2500
Germany (DAX)
France (CAC 40)
Europe (Stoxx 600)
$ 20 2000 Nether. (AEX)
Aus. (S&P/ASX 200)
Philippines (PSEi)
$ 15 1500 S.Africa (FTSE JSE T40)
China(Hang Seng)
UK (FTSE 100)
Spain (IBEX 35)
$ 10 1000
Canada (TSX)
Greece (ASE Athens)
Japan (TOPIX)
$5 500 S.Korea (KOSPI)
US (S&P 500)
Brazil (Ibovespa)
$ 0 Mexico (Mex IPC)




















-2% 2% 6% 10% 14% 18% 22% 26% 30%

Source: PricewaterhouseCoopers LLP/National Venture Capital Association Source: Bloomberg

MoneyTree Report; data by Thomson Reuters

Murmurings in Europe Spot the odd one out

MSCI EAFE index relative to S&P 500, Jan 2014 = 100 Employees in the US by industry, 2003 indexed to 100
MSCI EAFE / S&P 500 190
Oil and gas
Transportation &
170 Utilities

All industries

150 Mining

95 Wholesale trade

130 Retail trade

90 Government

Fin. Services




























Source: Datastream (EAFE - 'Europe, Australasia, Far East) Source: BLS, BEA

Goldman Sachs Global Investment Research 20

Fortnightly Thoughts Issue 85

Disclosure Appendix

Reg AC
We, Hugo Scott-Gall, Sumana Manohar, Alex Blostein, Greg Dunham, Yuichiro Isayama and Takafumi Hara, hereby certify that all of the
views expressed in this report accurately reflect our personal views about the subject company or companies and its or their securities. We
also certify that no part of our compensation was, is, or will be, directly or indirectly, related to the specific recommendations or views
expressed in this report.
Unless otherwise stated, the individuals listed on the cover page of this report are analysts in Goldman Sachs' Global Investment Research

Investment Profile
The Goldman Sachs Investment Profile provides investment context for a security by comparing key attributes of that security to its peer
group and market. The four key attributes depicted are: growth, returns, multiple and volatility. Growth, returns and multiple are indexed
based on composites of several methodologies to determine the stocks percentile ranking within the region's coverage universe.
The precise calculation of each metric may vary depending on the fiscal year, industry and region but the standard approach is as follows:
Growth is a composite of next year's estimate over current year's estimate, e.g. EPS, EBITDA, Revenue. Return is a year one prospective
aggregate of various return on capital measures, e.g. CROCI, ROACE, and ROE. Multiple is a composite of one-year forward valuation
ratios, e.g. P/E, dividend yield, EV/FCF, EV/EBITDA, EV/DACF, Price/Book. Volatility is measured as trailing twelve-month volatility
adjusted for dividends.

Quantum is Goldman Sachs' proprietary database providing access to detailed financial statement histories, forecasts and ratios. It can be
used for in-depth analysis of a single company, or to make comparisons between companies in different sectors and markets.

GS SUSTAIN is a global investment strategy aimed at long-term, long-only performance with a low turnover of ideas. The GS SUSTAIN
focus list includes leaders our analysis shows to be well positioned to deliver long term outperformance through sustained competitive
advantage and superior returns on capital relative to their global industry peers. Leaders are identified based on quantifiable analysis of
three aspects of corporate performance: cash return on cash invested, industry positioning and management quality (the effectiveness of
companies' management of the environmental, social and governance issues facing their industry).

Disclosure Appendix
Coverage group(s) of stocks by primary analyst(s)
There are no coverage groups associated with the analyst(s).
Company-specific regulatory disclosures
The following disclosures relate to relationships between The Goldman Sachs Group, Inc. (with its affiliates, "Goldman Sachs") and
companies covered by the Global Investment Research Division of Goldman Sachs and referred to in this research.
Compendium report: please see disclosures at Disclosures applicable to the companies included
in this compendium can be found in the latest relevant published research.
Distribution of ratings/investment banking relationships
Goldman Sachs Investment Research global coverage universe
Rating distribution Investment Banking Relationships

Buy Hold Sell Buy Hold Sell

Global 33% 54% 13% 44% 38% 32%

As of January 1, 2015, Goldman Sachs Global Investment Research had investment ratings on 3,483 equity securities. Goldman Sachs
assigns stocks as Buys and Sells on various regional Investment Lists; stocks not so assigned are deemed Neutral. Such assignments
equate to Buy, Hold and Sell for the purposes of the above disclosure required by NASD/NYSE rules. See 'Ratings, Coverage groups and
views and related definitions' below.

Goldman Sachs Global Investment Research 21

Fortnightly Thoughts Issue 85

Price target and rating history chart(s)

Compendium report: please see disclosures at Disclosures applicable to the companies included
in this compendium can be found in the latest relevant published research.

Regulatory disclosures
Disclosures required by United States laws and regulations
See company-specific regulatory disclosures above for any of the following disclosures required as to companies referred to in this report:
manager or co-manager in a pending transaction; 1% or other ownership; compensation for certain services; types of client relationships;
managed/co-managed public offerings in prior periods; directorships; for equity securities, market making and/or specialist role. Goldman
Sachs usually makes a market in fixed income securities of issuers discussed in this report and usually deals as a principal in these
The following are additional required disclosures: Ownership and material conflicts of interest: Goldman Sachs policy prohibits its
analysts, professionals reporting to analysts and members of their households from owning securities of any company in the analyst's area
of coverage. Analyst compensation: Analysts are paid in part based on the profitability of Goldman Sachs, which includes investment
banking revenues. Analyst as officer or director: Goldman Sachs policy prohibits its analysts, persons reporting to analysts or
members of their households from serving as an officer, director, advisory board member or employee of any company in the analyst's area
of coverage. Non-U.S. Analysts: Non-U.S. analysts may not be associated persons of Goldman, Sachs & Co. and therefore may not be
subject to NASD Rule 2711/NYSE Rules 472 restrictions on communications with subject company, public appearances and trading
securities held by the analysts.
Distribution of ratings: See the distribution of ratings disclosure above. Price chart: See the price chart, with changes of ratings and
price targets in prior periods, above, or, if electronic format or if with respect to multiple companies which are the subject of this report, on
the Goldman Sachs website at
Additional disclosures required under the laws and regulations of jurisdictions other than the United States
The following disclosures are those required by the jurisdiction indicated, except to the extent already made above pursuant to United
States laws and regulations. Australia: Goldman Sachs Australia Pty Ltd and its affiliates are not authorised deposit-taking institutions (as
that term is defined in the Banking Act 1959 (Cth)) in Australia and do not provide banking services, nor carry on a banking business, in
Australia. This research, and any access to it, is intended only for "wholesale clients" within the meaning of the Australian Corporations Act,
unless otherwise agreed by Goldman Sachs. In producing research reports, members of the Global Investment Research Division of
Goldman Sachs Australia may attend site visits and other meetings hosted by the issuers the subject of its research reports. In some
instances the costs of such site visits or meetings may be met in part or in whole by the issuers concerned if Goldman Sachs Australia
considers it is appropriate and reasonable in the specific circumstances relating to the site visit or meeting. Brazil: Disclosure information
in relation to CVM Instruction 483 is available at Where applicable, the Brazil-
registered analyst primarily responsible for the content of this research report, as defined in Article 16 of CVM Instruction 483, is the first
author named at the beginning of this report, unless indicated otherwise at the end of the text. Canada: Goldman Sachs Canada Inc. is
an affiliate of The Goldman Sachs Group Inc. and therefore is included in the company specific disclosures relating to Goldman Sachs (as
defined above). Goldman Sachs Canada Inc. has approved of, and agreed to take responsibility for, this research report in Canada if and to
the extent that Goldman Sachs Canada Inc. disseminates this research report to its clients. Hong Kong: Further information on the
securities of covered companies referred to in this research may be obtained on request from Goldman Sachs (Asia) L.L.C. India: Further
information on the subject company or companies referred to in this research may be obtained from Goldman Sachs (India) Securities
Private Limited. Japan: See below. Korea: Further information on the subject company or companies referred to in this research may be
obtained from Goldman Sachs (Asia) L.L.C., Seoul Branch. New Zealand: Goldman Sachs New Zealand Limited and its affiliates are
neither "registered banks" nor "deposit takers" (as defined in the Reserve Bank of New Zealand Act 1989) in New Zealand. This research,
and any access to it, is intended for "wholesale clients" (as defined in the Financial Advisers Act 2008) unless otherwise agreed by
Goldman Sachs. Russia: Research reports distributed in the Russian Federation are not advertising as defined in the Russian legislation,
but are information and analysis not having product promotion as their main purpose and do not provide appraisal within the meaning of the
Russian legislation on appraisal activity. Singapore: Further information on the covered companies referred to in this research may be
obtained from Goldman Sachs (Singapore) Pte. (Company Number: 198602165W). Taiwan: This material is for reference only and must
not be reprinted without permission. Investors should carefully consider their own investment risk. Investment results are the responsibility
of the individual investor. United Kingdom: Persons who would be categorized as retail clients in the United Kingdom, as such term is
defined in the rules of the Financial Conduct Authority, should read this research in conjunction with prior Goldman Sachs research on the
covered companies referred to herein and should refer to the risk warnings that have been sent to them by Goldman Sachs International. A
copy of these risks warnings, and a glossary of certain financial terms used in this report, are available from Goldman Sachs International
on request.
European Union: Disclosure information in relation to Article 4 (1) (d) and Article 6 (2) of the European Commission Directive
2003/126/EC is available at which states the European Policy for Managing Conflicts of
Interest in Connection with Investment Research.
Japan: Goldman Sachs Japan Co., Ltd. is a Financial Instrument Dealer registered with the Kanto Financial Bureau under registration
number Kinsho 69, and a member of Japan Securities Dealers Association, Financial Futures Association of Japan and Type II Financial
Instruments Firms Association. Sales and purchase of equities are subject to commission pre-determined with clients plus consumption tax.
See company-specific disclosures as to any applicable disclosures required by Japanese stock exchanges, the Japanese Securities
Dealers Association or the Japanese Securities Finance Company.
Ratings, coverage groups and views and related definitions
Buy (B), Neutral (N), Sell (S) -Analysts recommend stocks as Buys or Sells for inclusion on various regional Investment Lists. Being
assigned a Buy or Sell on an Investment List is determined by a stock's return potential relative to its coverage group as described below.
Any stock not assigned as a Buy or a Sell on an Investment List is deemed Neutral. Each regional Investment Review Committee manages
various regional Investment Lists to a global guideline of 25%-35% of stocks as Buy and 10%-15% of stocks as Sell; however, the
distribution of Buys and Sells in any particular coverage group may vary as determined by the regional Investment Review Committee.

Goldman Sachs Global Investment Research 22

Fortnightly Thoughts Issue 85

Regional Conviction Buy and Sell lists represent investment recommendations focused on either the size of the potential return or the
likelihood of the realization of the return.
Return potential represents the price differential between the current share price and the price target expected during the time horizon
associated with the price target. Price targets are required for all covered stocks. The return potential, price target and associated time
horizon are stated in each report adding or reiterating an Investment List membership.
Coverage groups and views: A list of all stocks in each coverage group is available by primary analyst, stock and coverage group at The analyst assigns one of the following coverage views which represents the analyst's
investment outlook on the coverage group relative to the group's historical fundamentals and/or valuation. Attractive (A). The investment
outlook over the following 12 months is favorable relative to the coverage group's historical fundamentals and/or valuation. Neutral (N).
The investment outlook over the following 12 months is neutral relative to the coverage group's historical fundamentals and/or valuation.
Cautious (C). The investment outlook over the following 12 months is unfavorable relative to the coverage group's historical
fundamentals and/or valuation.
Not Rated (NR). The investment rating and target price have been removed pursuant to Goldman Sachs policy when Goldman Sachs is
acting in an advisory capacity in a merger or strategic transaction involving this company and in certain other circumstances. Rating
Suspended (RS). Goldman Sachs Research has suspended the investment rating and price target for this stock, because there is not a
sufficient fundamental basis for determining, or there are legal, regulatory or policy constraints around publishing, an investment rating or
target. The previous investment rating and price target, if any, are no longer in effect for this stock and should not be relied upon.
Coverage Suspended (CS). Goldman Sachs has suspended coverage of this company. Not Covered (NC). Goldman Sachs does
not cover this company. Not Available or Not Applicable (NA). The information is not available for display or is not applicable. Not
Meaningful (NM). The information is not meaningful and is therefore excluded.
Global product; distributing entities
The Global Investment Research Division of Goldman Sachs produces and distributes research products for clients of Goldman Sachs on a
global basis. Analysts based in Goldman Sachs offices around the world produce equity research on industries and companies, and
research on macroeconomics, currencies, commodities and portfolio strategy. This research is disseminated in Australia by Goldman
Sachs Australia Pty Ltd (ABN 21 006 797 897); in Brazil by Goldman Sachs do Brasil Corretora de Ttulos e Valores Mobilirios S.A.; in
Canada by either Goldman Sachs Canada Inc. or Goldman, Sachs & Co.; in Hong Kong by Goldman Sachs (Asia) L.L.C.; in India by
Goldman Sachs (India) Securities Private Ltd.; in Japan by Goldman Sachs Japan Co., Ltd.; in the Republic of Korea by Goldman Sachs
(Asia) L.L.C., Seoul Branch; in New Zealand by Goldman Sachs New Zealand Limited; in Russia by OOO Goldman Sachs; in Singapore by
Goldman Sachs (Singapore) Pte. (Company Number: 198602165W); and in the United States of America by Goldman, Sachs & Co.
Goldman Sachs International has approved this research in connection with its distribution in the United Kingdom and European Union.
European Union: Goldman Sachs International authorised by the Prudential Regulation Authority and regulated by the Financial
Conduct Authority and the Prudential Regulation Authority, has approved this research in connection with its distribution in the European
Union and United Kingdom; Goldman Sachs AG and Goldman Sachs International Zweigniederlassung Frankfurt, regulated by the
Bundesanstalt fr Finanzdienstleistungsaufsicht, may also distribute research in Germany.
General disclosures
This research is for our clients only. Other than disclosures relating to Goldman Sachs, this research is based on current public information
that we consider reliable, but we do not represent it is accurate or complete, and it should not be relied on as such. We seek to update our
research as appropriate, but various regulations may prevent us from doing so. Other than certain industry reports published on a periodic
basis, the large majority of reports are published at irregular intervals as appropriate in the analyst's judgment.
Goldman Sachs conducts a global full-service, integrated investment banking, investment management, and brokerage business. We have
investment banking and other business relationships with a substantial percentage of the companies covered by our Global Investment
Research Division. Goldman, Sachs & Co., the United States broker dealer, is a member of SIPC (
Our salespeople, traders, and other professionals may provide oral or written market commentary or trading strategies to our clients and
our proprietary trading desks that reflect opinions that are contrary to the opinions expressed in this research. Our asset management area,
our proprietary trading desks and investing businesses may make investment decisions that are inconsistent with the recommendations or
views expressed in this research.
The analysts named in this report may have from time to time discussed with our clients, including Goldman Sachs salespersons and
traders, or may discuss in this report, trading strategies that reference catalysts or events that may have a near-term impact on the market
price of the equity securities discussed in this report, which impact may be directionally counter to the analyst's published price target
expectations for such stocks. Any such trading strategies are distinct from and do not affect the analyst's fundamental equity rating for such
stocks, which rating reflects a stock's return potential relative to its coverage group as described herein.
We and our affiliates, officers, directors, and employees, excluding equity and credit analysts, will from time to time have long or short
positions in, act as principal in, and buy or sell, the securities or derivatives, if any, referred to in this research.
This research is not an offer to sell or the solicitation of an offer to buy any security in any jurisdiction where such an offer or solicitation
would be illegal. It does not constitute a personal recommendation or take into account the particular investment objectives, financial
situations, or needs of individual clients. Clients should consider whether any advice or recommendation in this research is suitable for their
particular circumstances and, if appropriate, seek professional advice, including tax advice. The price and value of investments referred to
in this research and the income from them may fluctuate. Past performance is not a guide to future performance, future returns are not
guaranteed, and a loss of original capital may occur. Fluctuations in exchange rates could have adverse effects on the value or price of, or
income derived from, certain investments.
Certain transactions, including those involving futures, options, and other derivatives, give rise to substantial risk and are not suitable for all
investors. Investors should review current options disclosure documents which are available from Goldman Sachs sales representatives or
at Transaction costs may be significant in option strategies calling for multiple
purchase and sales of options such as spreads. Supporting documentation will be supplied upon request.
All research reports are disseminated and available to all clients simultaneously through electronic publication to our internal client
websites. Not all research content is redistributed to our clients or available to third-party aggregators, nor is Goldman Sachs responsible

Goldman Sachs Global Investment Research 23

Fortnightly Thoughts Issue 85

for the redistribution of our research by third party aggregators. For research, models or other data available on a particular security, please
contact your sales representative or go to
Disclosure information is also available at or from Research Compliance,
200 West Street, New York, NY 10282.
2015 Goldman Sachs.
No part of this material may be (i) copied, photocopied or duplicated in any form by any means or (ii) redistributed
without the prior written consent of The Goldman Sachs Group, Inc.

Goldman Sachs Global Investment Research 24